Many types of markets, the labor market among them, take place online over the internet. For that reason, leveraging the internet as a data source of social science, and labor economics in particular, is at the core of the research mission of IDSC, IZA’s research data center.
When markets take place online, the underlying information and communications technology is used to optimize the matching of supply and demand, which is the core problem of any market. Generating transaction data in such markets can be done easily and efficiently, due to the nature of digital technology, which can then be used to rewind and replay the markets so that studying and understanding such markets depends heavily on access to such data.
Organized by Nikos Askitas and Peter J. Kuhn, the fourth IDSC workshop brought together economists from across the globe to showcase research with online data in three sessions (Search and Matching / Evidence from Vacancy Postings / Search Behavior), featuring three keynote lectures (“Salary History and Employer Demand” / “Online Jobs Vacancies in the Covid Crisis” / “Algorithmic Hiring”). A key feature of this year’s workshop was the large number of high quality, deep, multidisciplinary job market papers, some of which are summarized below.
Solving congestion in labor market recommender systems
Designing labor market recommender systems is a fundamentally different problem than designing product recommender systems. This is because most workers can only work for a single firm, while firms can often serve millions of customers at the same time. Thus, while it makes sense for algorithms to suggest the same movie to a large number of customers, suggesting the same worker to all firms would be disastrous. Using data from the French Public Employment Service (PES), Bruno Crépon and co-authors propose and evaluate improvements to existing algorithms that solve this ‘congestion’ problem. Using the mathematics of optimal transport, the study generates substantial improvements on product-type labor market recommendation algorithms.
Predicting wage premia from the language of job postings
Following the rich tradition in the economics literature of estimating wage premia for various job characteristics by applying hedonic regression, Sarah H. Bana applied natural language processing (NLP) techniques on data with salary information from Greenwich.HR linked with job postings data from Burning Glass Technologies to build a model that predicts salaries from job postings text. The model explains 73% of the variation, which is 10 percentage points above a fixed effects model using occupation and location. The result of the paper is a crucial input in the matching process as firms and workers make strategic decisions in the two-sided market.
Alma mater matters: A global look at university quality
While several countries have popular ranking systems for their universities, to date it has been hard to compare the value of degrees from universities in different countries. Using Glassdoor data on the earnings of a college’s graduates, and exploiting the fact that graduates from top universities are increasingly internationally mobile, Jason Sockin and co-authors are able to solve this comparability problem. While their ranking of colleges is correlated with existing rankings (such as U.S. News), it ranks liberal arts colleges and top science and engineering schools in developing countries much higher. Their paper shows that graduates of the latter schools make an outsized contribution to world entrepreneurship and innovation, regardless of the country they work in.
For a full list of presentations see the workshop program.